Towards unified aleatory and epistemic uncertainty quantification for machinery health prognostic through sequential heteroscedastic Gaussian process regression

Predicting the Remaining Useful Life (RUL) is crucial for preventing catastrophic failures in industrial manufacturing. Traditional approaches often overlook a comprehensive quantification of the uncertainties associated with equipment degradation, impacting the predictions' reliability and clarity. This paper proposes an advanced uncertainty-aware machinery health prognostic algorithm, designed to cohesively measure both aleatory and epistemic uncertainties within a single framework. A Monotonic Recurrent Autoencoder (MRAE), enhanced with innovative penalty terms, is developed to accurately depict both the local and overall degradation patterns of machinery. Following this, a Sequential Heteroscedastic Gaussian Process Regression (SHGPR) model is proposed. This model employs two distinct Gaussian Processes (GPs) to assess the spatio-temporal aspects of the uncertainties related to equipment degradation. Leveraging Bayesian theory allows these GPs to be integrated seamlessly, facilitating a uniform approach to quantify the uncertainties from both sources. Utilizing the degradation data, this method enables precise inference of the RUL distribution. The effectiveness and feasibility of the proposed algorithm are demonstrated through its application to real-world mineral processing equipment.